Manual scale doesn't work if hpaSpec is set

[stingshen]

Describe the bug

Manual scale doesn't work if hpaSpec is set.

To reproduce

Apply the following yaml to k8s:

apiVersion: machinelearning.seldon.io/v1
kind: SeldonDeployment
metadata:
  name: seldon-model
spec:
  name: test-deployment
  replicas: 5
  predictors:
  - componentSpecs:
    - hpaSpec:
        maxReplicas: 3
        metrics:
        - resource:
            name: cpu
            targetAverageUtilization: 10
          type: Resource
        minReplicas: 1
      replicas: 2
      spec:
        containers:
        - image: seldonio/mock_classifier_rest:1.3
          imagePullPolicy: IfNotPresent
          name: classifier
          resources:
            requests:
              cpu: '0.2'
        terminationGracePeriodSeconds: 1
    graph:
      children: []
      endpoint:
        type: REST
      name: classifier
      type: MODEL
    name: example

Manual replica setting doesn't work. There is only one replica created.

Expected behaviour

There are 2 replicas created.

Environment

seldon 1.3.0 with K8s 1.18.4

[ukclivecox]

Are you saying for a Deployment + HPA with above setting if you try to manually set replicas to 5 it will go to 3? In this case I would expect maybe 2 as you set that for replicas in predictor?

[stingshen]

Sorry, I wrote the wrong expected replica number. The expected behavior is 2 replicas are created. The actual behavior is there is only one replica created by seldon deployment controller.

[ukclivecox]

We are following the advice here kubernetes/kubernetes#25238 (comment)
Replica setting is ignored when an HPA is applied. With normal Deployments you should not set replicas when using an HPA.

Please reopen if you think this is not correct.